I would say that you should look at the language's theoretical performance as much as its actual performance. A lot of these benchmarks are just a result of someone putting in a lot of time tweaking things until they are perfectly suited to the specific benchmark, in real-life you will get nowhere near that. Additionally, you won't actually need to get anywhere near it. If you do, then you will have the time to tweak things and get to know the language inside out. In that case, all that will matter is the theoretical performance of that language.

So what do I mean by theoretical performance? Well, Nim is a statically typed and compiled language, it stands to reason that these attributes make it very performant indeed. Even if certain aspects of the language are not efficient right now, you can bet that it will be fairly trivial to improve their performance. For languages like Python, Ruby, etc. you are heavily limited by the dynamic typing (which restricts the amount of information you have about the code) and the fact that they are primarily interpreted languages. You will need to put far more effort into optimising them than compiled languages.

As a side note, I do intend to improve the performance of Jester as well. But I'm not the only one doing this, the mofuw framework is #24 on the TechEmpower benchmarks right now.[1]

1 - https://twitter.com/nim_lang/status/1004849910699712513